Recently I lowered my goals, but I ended up sticking with it for longer. In the past, I kept thinking I had to catch every trend and fully understand every on-chain data point—only to find that the more I stared, the more anxious I became. Now I just look at the inflows and outflows of one or two protocols I care about each day, and occasionally jot down my impressions on the fly—somehow, I can even spot things I used to overlook.

Speaking of AI agents and on-chain interaction, I’ve been observing several automation scripts that do MEV. They do run pretty fast, but when a contract has just been upgraded or the ordering rules get changed temporarily, you still need someone to watch and adjust the parameters. Basically, logical gaps can be patched with code, but the judgment of whether to intervene at this particular moment still needs a human to catch it in the loop.

Lately, when looking at on-chain data, I’ve confirmed that the structure of validator revenue is indeed changing, and there are also plenty of complaints from retail users about unfair ordering. To be honest, many automated strategies are essentially exploiting human weaknesses, and you can’t fully model human nature in code. Lower the expectations first, and you’re actually more able to see which steps are truly critical—where the humans are.